OpenAI image API pricing is easy to misread if you treat every image request as one flat "cost per image." GPT Image pricing is token-based: text prompt tokens, image input tokens for edits or references, cached input when available, image output tokens, and in some GPT Image rows, text output tokens. Resolution, quality, API path, and whether you are generating or analyzing images all change the bill.
As of June 11, 2026, OpenAI's pricing page lists current GPT image-family rows for gpt-image-2, gpt-image-1.5, and gpt-image-1-mini. OpenAI's image generation guide still includes examples for gpt-image-1, but teams budgeting new work should check the current pricing table before copying older GPT Image or DALL-E cost assumptions.
For teams using Flatkey, the practical answer is: use OpenAI's pricing page to understand direct-provider mechanics, then use Flatkey's pricing page and usage dashboard to audit routed usage for the exact model row your application calls.
Quick Answer: What OpenAI Image API Pricing Includes
OpenAI image API pricing has four cost buckets to check, though not every model row uses all of them:
| Cost Bucket | When It Applies | Budget Variable | What To Verify |
|---|---|---|---|
| Text input tokens | Your prompt, instructions, and any text context sent with the request. | Prompt length, templates, hidden system instructions, and repeated context. | The current text input rate for the image model on OpenAI pricing. |
| Text output tokens | Model-generated text returned by rows that list a text output price. | Whether the workflow asks the model to return text, structured output, or captions alongside image work. | The model-specific text output row; for example, gpt-image-1.5 lists a text output rate. |
| Image input tokens | Image edits, reference images, masks, and multimodal image analysis. | Input image count, image size, detail setting, and fidelity requirements. | The model-specific image input rate and image tokenization rules. |
| Image output tokens | Generated images. | Number of images, resolution, quality, and output model. | The model-specific output rate and output-cost examples for your size and quality. |
The simple budgeting formula is:
estimated cost =
text input tokens x text input rate
+ optional text output tokens x text output rate
+ image input tokens x image input rate
+ image output tokens x image output rate
That formula is why OpenAI image API pricing changes so quickly between a short text-to-image prompt, a high-quality image edit with two references, and a multimodal workflow that analyzes an uploaded image before generating a new one.
Current GPT Image Price Rows To Check
OpenAI prices GPT image models per 1M tokens. The table below summarizes the direct OpenAI standard API rows visible on June 11, 2026. Treat this as a publish-day snapshot, not a permanent contract.
| OpenAI Model | Text Input | Cached Text Input | Text Output | Image Input | Cached Image Input | Image Output | Use In Budgeting |
|---|---|---|---|---|---|---|---|
gpt-image-2 |
$5 / 1M tokens | $1.25 / 1M tokens | Not listed | $8 / 1M tokens | $2 / 1M tokens | $30 / 1M tokens | Current flagship GPT Image generation and editing budgets. |
gpt-image-1.5 |
$5 / 1M tokens | $1.25 / 1M tokens | $10 / 1M tokens | $8 / 1M tokens | $2 / 1M tokens | $32 / 1M tokens | Compare only if your account or endpoint is actually using this row. |
gpt-image-1-mini |
$2 / 1M tokens | $0.20 / 1M tokens | Not listed | $2.50 / 1M tokens | $0.25 / 1M tokens | $8 / 1M tokens | Lower-cost image generation experiments and high-volume drafts. |
OpenAI also lists Batch API pricing at roughly half the standard token rates for these image rows. That can matter for asynchronous image jobs, backfills, test suites, and content pipelines that do not need an immediate response.
Example Output Costs Per Generated Image
The output portion of OpenAI image API pricing changes by model, size, and quality. OpenAI's image generation guide gives calculator examples for generated image output. For a square 1024 x 1024 image, the current examples are:
| Model | Low Quality | Medium Quality | High Quality | Budget Meaning |
|---|---|---|---|---|
gpt-image-2 |
$0.006 | $0.053 | $0.211 | Quality is the biggest lever when output volume is high. |
gpt-image-1.5 |
$0.009 | $0.034 | $0.133 | Check whether this model is actually used before budgeting from it. |
gpt-image-1-mini |
$0.005 | $0.011 | $0.036 | Useful for lower-cost drafts before upgrading final renders. |
gpt-image-1 example |
$0.011 | $0.042 | $0.167 | Legacy reference point for older GPT Image pricing content. |
For a quick volume estimate, multiply the output example by request count, then add text input, image input, and any billable text output costs. One thousand square gpt-image-2 medium-quality outputs would be about $53 in image output cost before prompt and input-image costs. The same 1,000 outputs at high quality would be about $211 before input costs. That spread is why a serious OpenAI image API pricing budget should separate draft, preview, and final-quality traffic.
Generation, Editing, And Vision Are Different Budgets
Teams often mix three different cost questions into one OpenAI image API pricing search:
| Workflow | What You Are Buying | Cost Driver | Budgeting Note |
|---|---|---|---|
| Text-to-image generation | A new image generated from text. | Text input plus image output tokens. | Start with output examples, then add prompt tokens. |
| Image editing | A new or modified image generated from one or more input images. | Text input, image input, and image output tokens. | Reference images and masks add input cost. High-fidelity inputs can cost more. |
| Vision or image analysis | A text model reads an uploaded image and returns text or structured output. | Image input tokens plus normal model output tokens. | Do not use image-generation output prices for pure image analysis. |
| Generate after analysis | A multimodal request analyzes an image, then generates another image. | Both analysis tokens and image-generation tokens. | Track these as two stages, even if your product hides them behind one button. |
The older search phrase "GPT-4 Vision API pricing" can be misleading here. Current OpenAI docs describe image inputs as tokenized inputs in multimodal model calls. Detail level, image dimensions, and model-specific token accounting determine image input cost. That is a different budget line from GPT Image output cost.
Budget Worksheet For GPT Image Features
Use this worksheet before shipping a feature that depends on OpenAI image API pricing.
| Decision | Default To Start | When To Raise Cost | What To Log |
|---|---|---|---|
| Model | Pick the exact model row, such as gpt-image-2 or gpt-image-1-mini. |
Use the higher-cost model for final outputs, brand-critical visuals, or quality-sensitive editing. | Model ID, endpoint, and provider path. |
| Quality | Use low or medium for previews and drafts. | Use high only when visual fidelity affects conversion, review, or production quality. | Quality setting and output size. |
| Resolution | Generate only the dimensions the product needs. | Increase size for final assets, not every intermediate attempt. | Requested size and final downloaded size. |
| Input images | Keep references and masks minimal. | Add references when edits need identity, layout, or style continuity. | Input image count, detail/fidelity, and retry count. |
| Batching | Use real-time calls for user-facing flows. | Use Batch API for delayed jobs when the workflow allows it. | Batch size, completion rate, and rejected jobs. |
| Retries | Cap automatic retries and show clear failure states. | Allow extra attempts only for paid workflows or internal production jobs. | Retries per output and cost per successful asset. |
This turns OpenAI image API pricing from a static table into an operating model: you know which settings belong in prototypes, which settings belong in paid user flows, and which settings require approval.
How Flatkey Helps Audit Multimodal Spend
Flatkey does not replace the need to understand direct OpenAI pricing. It gives teams one place to route and audit usage across model families. Flatkey's current public copy says teams can access Claude, GPT, Gemini, DeepSeek, Qwen, Seedance 2.0, GPT Image, and more with one API key, with clear pricing, unified billing, and one dashboard for keys, usage, and routing.
That matters for OpenAI image API pricing because image workloads are rarely isolated. A product may analyze an uploaded screenshot, generate a draft image, call a text model for captions, and route fallbacks through another provider. A unified usage view helps you see the blended cost instead of debugging invoices across separate provider accounts.
When you use Flatkey for image or multimodal routing, check these items:
- Open Flatkey pricing and verify the exact image model row before launch.
- Confirm the model ID your app calls, especially if you see both current and legacy GPT Image names.
- Run a small test batch and compare expected request count with usage logs.
- Track output size, quality, retries, and edit/reference-image count in your own product telemetry.
- Review costs after the first real user cohort, not only during a happy-path test.
In a live Flatkey pricing snapshot on June 11, 2026, the pricing endpoint returned 653 model rows and supported endpoint families including image-generation, openai, openai-response, and openai-video. It also included GPT image-family rows such as gpt-image-2, openai/gpt-image-2, gpt-image-1.5, gpt-image-1-mini, and legacy gpt-image-1. Use that as a reminder to verify the publish-day catalog row instead of assuming one stale model name.
Common Mistakes In OpenAI Image API Pricing Estimates
| Mistake | Why It Breaks The Budget | Better Practice |
|---|---|---|
| Using one average price per image | It hides model, quality, size, and edit-input differences. | Estimate by workflow type: draft, edit, final render, analysis, and batch. |
Copying old gpt-image-1 or DALL-E pricing into a new feature |
Current pricing rows and model availability can differ. | Check OpenAI pricing and your routed provider catalog on the deployment date. |
| Ignoring input images | Image edits and references add image input tokens. | Log the number of images attached to each request. |
| Letting retries run uncapped | Failed or low-quality attempts can become the hidden cost center. | Set retry limits and review cost per accepted output. |
| Mixing vision analysis with image generation | The input-token budget and output-image budget are not the same thing. | Track analysis and generation as separate stages. |
Final Checklist Before You Ship
- Pick the model row you will actually call and write down the exact model ID.
- Separate text-to-image, edit, analysis, and generate-after-analysis workflows.
- Estimate output cost at the specific quality and size you will use in production.
- Add text input, optional text output, and image input costs for prompts, references, masks, or uploaded images.
- Decide whether any delayed jobs can use Batch API pricing.
- Set retry limits, quota limits, and usage alerts before real traffic begins.
- Use Flatkey pricing and usage logs to compare expected and actual routed spend.
The best OpenAI image API pricing plan is not a single number. It is a short operating loop: verify the current model row, estimate by workflow, run a measured test batch, and review real usage in one dashboard before opening the feature to more users.
View Pricing to check current Flatkey image model rows and compare multimodal usage before you scale a GPT Image workflow.
